What Cisco learned on its AI journey

Sales staff preparing for customer visits with AI assistance, an AI that supports customer service — these are just two examples of how Cisco Germany uses artificial intelligence internally in its daily work. But getting to this point was a long journey for the technology giant, fraught with numerous challenges.

Prompting — the new Googling

When Cisco Germany launched its AI journey, understanding and use of AI varied greatly among employees, says as Detlev Kühne, who, as managing director of partner sales and a member of the management board, was responsible for the internal AI implementation at Cisco Germany.

On a scale of 0 to 10, where 10 represents comprehensive knowledge and daily use of all important AI tools, most employees fell into the lower to middle range, he says. “When you actually ask what AI even means, you realize just how different the knowledge bases are,” he adds.

Cisco responded by providing basic training for all employees. The training not only imparted knowledge about various AI platforms beyond ChatGPT, but also placed a strong emphasis on the art of effective prompting. For Kühne, prompting is “the new Google,” and its quality significantly influences the AI’s output.

Furthermore, Cisco’s training included basic information on the legal framework for AI use, particularly with regard to GDPR and the EU AI Act. “Every employee who feeds AI anywhere should know what they are doing. And they should also consider how to handle the data they receive as output,” Kühne emphasizes.

The employees’ self-assessment before the training showed that the majority rated themselves between 2.5 and 3 on a scale of 0 to 5. Surprisingly, many older employees, so-called “silverbacks,” actively contributed to the discussion about AI. This refuted the prejudice that AI is primarily a topic for the younger generation, even though many younger colleagues at Cisco were already using AI extensively — for example, for vacation planning, email drafting, or summarizing client meetings.

Circuit as its own LLM

Internal AI usage at Cisco Germany really took off in 2024, when Cisco launched its in-house AI application Circuit in May of that year.

Circuit is an evolution of Bridge IT (announced in February 2024) and Enterprise Chat AI (in use since August 2022). With the launch of Circuit, the number of regular AI users increased dramatically worldwide. Around 50,000 of Cisco’s more than 80,000 employees now use AI in their work.

According to Kühne, Circuit is more than just another AI tool; it’s a secure, proprietary AI platform based on Cisco’s own large language model (LLM). Circuit directly integrates into Cisco’s collaboration tool Webex, which is used internally as the standard chat tool. Alternatively, the AI ​​platform can be accessed via a browser version.

For Kühne, using an in-house AI offers a crucial advantage: “Because we host the AI ​​on our own servers, I can essentially put all the data into it.” Ultimately, the data remains internal and is protected up to the second-highest security level, which alleviates employees’ concerns about whether they are allowed to enter sensitive information. “Everything stays with us; nothing goes out anywhere. It’s all completely isolated,” Kühne emphasizes.

Shadow AI challenge

Nevertheless, Cisco — like other companies — faces the challenge of shadow AI, in which employees, out of convenience or ignorance, use external, insecure tools such as ChatGPT for business purposes. Cisco addresses this, among other things, with regular training and certifications that emphasize the proper handling of confidential data.

In addition, Cisco relies on technical solutions such as its in-house security platform, AI Defense. This product monitors employees’ use of AI to prevent misuse of personal or company data. According to Kühne, this is particularly important in development, where tools such as GitHub Copilot are often used. Kühne also recommends this three-pronged approach of trust, awareness, and technical control to other companies.

AI outputs need attention

Another challenge for Cisco remains the quality of AI outputs and the risk of the much-discussed AI hallucinations. Cisco has gained a clear insight in this regard: cross-referencing and verifying AI outputs is essential, even if Circuit, by using proprietary data, delivers less inaccurate information.

Kühne explains that external AI systems sometimes recommended using long-discontinued products in their responses, instead of mentioning current devices. “Not everything an AI outputs can be used safely,” Kühne warns.

Internal AI use cases at Cisco

Cisco’s AI strategy is fundamentally “Start Small, Then Go Big.” In other words, the goal is to use AI effectively where it offers direct added value — regardless of the application’s size. According to Kühne, the focus is primarily on sales and customer service (CX).

In sales, employees use AI for various purposes, including:

• In-depth customer research: Before calling or visiting a customer, account managers can gather detailed information about the customer’s core business and AI usage to develop tailored offers.
• Communication: Assistance with writing emails, summarizing customer appointments, or quickly answering customer questions during a phone call.
• Preparing for customer meetings: Employees download the latest customer news as AI-generated podcasts and listen to them on their way to the meeting, instead of searching websites.
• Technical assurance: Technical sales colleagues use AI to cross-check product details before sending customer responses.

In post-sales and CX, AI relieves the burden on employees by:

• Automated case resolution: Already 25% of CX cases are solved by AI.
• Reduced processing times: AI helps to significantly reduce the time required to answer support requests.
• Employees freed from routine inquiries: This allows support staff to concentrate on more complex tasks. However, the human contact person remains the first point of contact externally.

5 lessons learned

When asked what recommendations he has for other decision-makers from Cisco’s AI journey, Kühne has five key lessons to offer:

• Identify the “fours and fives”: It is important to know who is already enthusiastically using AI in order to win these employees over as multipliers for others
• Involve managers as pacesetters: “The manager is the one who sets the pace,” says Kühne. If leaders ignore AI, their teams will too. Managers must lead by example, share their own experiences — including failures with prompts — and create an environment that encourages experimentation.
• Foster regular knowledge transfer: A one-off training session is not enough. Continuous formats such as team meetings or informal gatherings are needed, where employees can exchange their experiences in a safe space.
• Initial assessment is key: Companies should first determine what their employees are already doing with AI. Many managers will be surprised by how much AI is already being used, albeit hidden.
• Make AI a top priority: Unlike the cloud, which was not necessarily a “top priority” at the beginning, AI must be supported by company management and function as a partnership between IT and the business units in order to be filled with relevant content.

Kühne also notes self-critically: “I started the AI ​​journey far too late.”

He therefore advises other leaders not to make the same mistake. And the frequently encountered fear of AI often stems from a lack of knowledge. This can only be dispelled through communication, examples, and demonstrating the possibilities.

Ultimately, AI should become an integral part of corporate culture. “Because the cloud may disappear again, but AI is here to stay,” Kühne concludes.